Why we rank skills by runs, not votes

Every directory eventually fills with content optimized for the directory instead of the user. Upvotes reward thumbnails and titles. Star ratings reward asking your audience to rate. We wanted a signal that is expensive to fake and cheap to earn by being genuinely useful, so skills here are ranked by one thing: how often they actually run.

How a run is counted

A run is counted when an agent fetches a skill through our MCP endpoint or someone installs it via the CLI. Nobody clicks a like button. If a skill keeps getting fetched week after week, that is because it keeps working on other people's footage.

Results close the loop

After running a skill, you can report back:

npx editfork report <slug> --worked
npx editfork report <slug> --broke

Reports are shown on the skill page, with links to what people made when they include one. A skill with a thousand runs and a pile of "broke" reports ranks worse than a smaller skill that reliably works. When Resolve or a model updates and a skill silently breaks, the results flip and everyone can see it.

What this means for creators

There is no growth hack here. You cannot buy placement (MCP replies will never contain sponsored results), and self-votes do not exist. The way to rank is to publish a skill that survives on footage you have never seen. That is the whole game, and we think it is the right game.

Skills live on the leaderboard, weekly and all-time. Publish yours with npx editfork share.